The purpose of this study was to develop a prognostic model for the survival of pediatric patients with rhabdomyosarcoma (RMS) using parameters that are measured during routine clinical management.
Yang et al BMC Cancer 2014, 14:654 http://www.biomedcentral.com/1471-2407/14/654 RESEARCH ARTICLE Open Access Prognostic model for predicting overall survival in children and adolescents with rhabdomyosarcoma Limin Yang1,2*, Tetsuya Takimoto1 and Junichiro Fujimoto1 Abstract Background: The purpose of this study was to develop a prognostic model for the survival of pediatric patients with rhabdomyosarcoma (RMS) using parameters that are measured during routine clinical management Methods: Demographic and clinical variables were evaluated in 1679 pediatric patients with RMS registered in the Surveillance, Epidemiology, and End Results (SEER) program from 1990 to 2010 A multivariate Cox proportional hazards model was developed to predict median, 5-year and 10-year overall survival (OS) The Akaike information criterion technique was used for model selection A nomogram was constructed using the reduced model after model selection, and was internally validated Results: Of the total 1679 patients, 543 died The 5-year OS rate was 64.5% (95% confidence interval (CI), 62.1-67.1%) and the 10-year OS was 61.8% (95%CI, 59.2-64.5%) for the entire cohort Multivariate analysis identified age at diagnosis, tumor size, histological type, tumor stage, surgery and radiotherapy as significantly associated with survival (p < 0.05) The bootstrap-corrected c-index for the model was 0.74 The calibration curve suggested that the model was well calibrated for all predictions Conclusions: This study provided an objective analysis of all currently available data for pediatric RMS from the SEER cancer registry A nomogram based on parameters that are measured on a routine basis was developed The nomogram can be used to predict 5- and 10-year OS with reasonable accuracy This information will be useful for estimating prognosis and in guiding treatment selection Keywords: Rhabdomyosarcoma, Cancer, Nomogram, Overall survival Background Rhabdomyosarcoma (RMS) is the most common softtissue sarcoma in children and adolescents and accounts for 3% of all pediatric tumors [1] Approximately 350 children are diagnosed with RMS in the United States every year [2] Incidence peaks at a very young age Because RMS is derived from immature striated skeletal muscle, this disease can occur at any site in the body Prognosis of RMS has improved significantly, with multidisciplinary management accounting for most of the increase in survival rate Since 1972, the Intergroup Rhabdomyosarcoma Study Group (IRSG) has conducted * Correspondence: yo-r@ncchd.go.jp Epidemiology and Clinical Research Center for Children’s Cancer, National Center for Child Health and Development, 2-10-1 Okura, Setagaya-ku, Tokyo 157-8535, Japan Division of Allergy, Department of Medical Subspecialties, Medical Support Center for Japan Environment and Children’s Study (JECS), National Center for Child Health and Development, 2-10-1 Okura, Setagaya-ku, Tokyo 157-8535, Japan a series of clinical trials and published a series of treatment guidelines for different primary sites As a result, the long-term survival rate of these patients has nearly tripled from approximately 25% in 1970 to more than 70% in the 1990s [3,4] The rarity of this disease means that most information regarding survival is derived from these clinical trials However, overall survival (OS) results differ between clinical trials and population-based cancer registries because of important differences between patients treated in routine practice and those treated in clinical trials For example, IRSG reports showed a 5-year OS of 70% in the 1990s [3,4], while, even in the 2000s, the 5-year OS was only approximately 50% in children with RMS according to population-based data [5] Clinical trials may select participants based on strict inclusion criteria, which consider the extent of disease, previous history of treatment, comorbidities, psychosocial conditions and © 2014 Yang et al.; licensee BioMed Central Ltd This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated Yang et al BMC Cancer 2014, 14:654 http://www.biomedcentral.com/1471-2407/14/654 Page of other factors [6]; patients in a poor condition may thus be excluded from the protocol OS in trials may therefore not reflect the prognosis of patients who receive treatment in a community setting Individualized estimation of the prognosis could be useful for counseling cancer patients on treatment selection and for optimizing therapeutic approaches [7] However, to the best of our knowledge, there is currently no such estimation tool for RMS based on patients from the general population In this study, we analyzed the OS in children and adolescents with RMS using population-based data collected by the Surveillance, Epidemiology, and End Results (SEER) program of the National Cancer Institute (NCI) [5], and constructed a nomogram based on variables collected from the routine cancer registry, with the aim of providing clinicians and patients with a practical clinical tool to predict survival Methods Study population The data were derived from the SEER program, which collects demographic, diagnostic and treatment information on all newly diagnosed cancer patients residing within specific US geographic regions Registry data are submitted without personal identifiers to the NCI, and these data are publicly available for research purpose Because all information in public-use SEER database remains de-identified, approval by an ethics committee was not necessary to perform the analysis [8] All authors have signed the data-use agreement and got permission from SEER program to use this data Using the SEER registry public database, we identified patients with RMS diagnosed from 1990 to 2010 [5] Children diagnosed with malignant, first primary RMS and aged 0–19 years were eligible for this analysis In this study, eligible RMS cases had International Classification of Childhood Cancer (ICCC) code IXa, corresponding to ICO-O-3 morphology codes: 1) RMS not otherwise specified 8900/3; 2) pleomorphic RMS adulttype 8901/3; 3) mixed-type RMS 8902/3; 4) embryonal RMS 8910/3; 5) spindle cell RMS 8912/3; 6) alveolar RMS 8920/3; or 7) embryonal sarcoma 8991/3 Patients were excluded from the analysis if the diagnosis was made at autopsy or by death certificate only Patients with no confirmation of diagnosis by microscopy were also excluded After selection, there were 1679 cases left in the cohort The flow chart for data selection is shown in Figure Data analysis In the description of variables and calculation of OS, age at diagnosis was classified as 0–4, 5–9, 10–14 or 15–19 years Age at diagnosis was treated as a continuous variable in multivariate analysis Other clinical factors Figure Flow chart for the creation of the Surveillance, Epidemiology, and End Results (SEER) data set included primary tumor site, histologic tumor subtype, tumor stage, tumor size, surgery and radiotherapy (RT) Primary tumor sites were classified as favorable or unfavorable based on the criteria for staging of pediatric tumors [9] The head and neck (nonparameningeal), genitourinary (non-bladder/prostate), and bile duct regions were defined as favorable sites, all other sites were defined as unfavorable, and an unknown site was regarded as a missing value Histology was classified as embryonal, alveolar or other histological subtype Histological subtypes with RMS not otherwise specified were treated as missing values Tumor stage was classified according to the SEER historic staging system Cases with insufficient information to define the stage were regarded as having a missing value Tumor size was truncated at 20 cm and was grouped into three levels for both character description and calculation of OS: 1) 0–4 cm; 2) 5–9 cm; and 3) ≥10 cm Size was treated as a continuous variable in the multivariate model Statistical methods All missing values were imputed with the ‘transcan’ function of the rms package [10] OS was calculated by the Kaplan-Meier product-limited method Survival curves were compared using the log-rank test Cox proportional hazard regressions were performed to assess the effects of covariates on OS For continuous variables, we fitted restricted cubic splines with three knots at 10%, 50% and 90% empirical quantiles We also considered the interaction effect between surgery and RT The proportional hazard assumption was justified by examining residual plots The Akaike information criterion was Yang et al BMC Cancer 2014, 14:654 http://www.biomedcentral.com/1471-2407/14/654 utilized for model selection We constructed a nomogram with the beta coefficients of variables in the reduced model The model was internally validated We generated 200 bootstrap samples to determine the calibration and discrimination of the model Calibration refers to the ability of a model to make unbiased estimates of outcome Calibration was assessed using a calibration curve generated by plotting the model-predicted 5-year and 10-year survival probabilities against the observed probability, as calculated by the Kaplan-Meier method The prognostic accuracy of the model was quantified by computing the concordance index (c-index) described by Harrell et al [11] The c-index is a discrimination measure that estimates the probability that, of two randomly chosen patients, the patient with the higher predicted survival will outlive the patient with the lower predicted survival The c-index ranges from 0.5 (no discrimination) to 1.0 (perfect discrimination) All statistical analyses were conducted using R version 3.1.0 software (Institute for Statistics and Mathematics, Vienna, Austria; www.r-project.org) [12] The model and nomogram were constructed using the R package rms [10] All statistical tests were two-sided, and values of p < 0.05 were considered significant Results Patient demographics are listed in Table A total of 1679 pediatric patients with RMS were included in the study Approximately 38.1% of the subjects were aged 0–4 years, 23.2% were 5–9 years, 20.6% were 10–14 years and 18.1% were 15–19 years There were 974 (58.0%) boys, and 705 (42.0%) girls The majority of patients were white (75.9%) Approximately 61.1% of RMS occurred at unfavorable sites Around 59.0% of patients were diagnosed with embryonal RMS, 33.2% with alveolar RMS and 7.7% with other RMS Based on SEER staging, 33.4% of patients had localized tumors, 34.9% had regional RMS and 31.7% had metastasis More than half (62.8%) of the patients had received RT, and 59.1% received surgery The 5-year OS rate for the entire cohort was 64.5% (95% confidence interval (CI), 62.1-67.1%) and the 10year OS rate was 61.8% (95%CI, 59.2-64.5%) Five- and 10-year OS rates by characteristic are listed in Table Sex and race had no influence on OS Prognosis worsened with increasing age; young children (0–4 years) had a better prognosis than adolescents (15–19 years), with 5-year OS of 71.3% and 47.9%, respectively Children with embryonal RMS had a longer survival than those with alveolar RMS, with estimated 5-year OS of 73.5% and 46.3%, respectively Patients with localized tumors had a better prognosis (5-year OS of 84.0%) than those with regional disease (72.4%) or distant metastasis Page of (35.7%) RMS at favorable sites had a better prognosis than that at unfavorable sites (p < 0.001) Patients with surgery had improved survival compared with those without surgery (p < 0.001) RT showed a weak but significant association with prognosis; 5-year OS was 65.6% in patients with RT compared with 62.7% in those without RT (p = 0.045) Multivariate analysis was performed using a Cox proportional hazards regression model We pre-specified nonlinearity for age at diagnosis and tumor size variables, and considered the effect on prognosis of the interaction between surgery and RT Residual plots indicated that the proportional hazards assumption held After model selection, we obtained a reduced model Beta coefficients and hazard ratios of variables are listed in Table The nomogram included age at diagnosis, size, tumor site, stage, histological type, surgery and RT (Figure 2) To use the nomogram, we drew a vertical line to the point row to assign point values for each variable, summed the point values for each variable to obtain total points, and then dropped a vertical line from the total points row to get the 5- and 10-year OS rates The model was internally validated Discrimination suggested good accuracy with a bootstrap-corrected cindex of 0.74, which denotes 74% probability that, of two randomly selected patients, the patient who survives longer will have a higher survival probability than the patient with shorter survival The calibration plots for 5and 10-year OS are shown in Figure Points in the calibration plot were close to the 45° line, which suggested that the model was well-calibrated for all predictions Discussion The current study evaluated OS among pediatric patients with newly-diagnosed RMS in a population-based dataset, and constructed a nomogram to predict 5- and 10-year OS This prognostic tool will be useful for estimating prognosis and guiding treatment selection The rarity of this disease means that most published studies are retrospective analyses of clinical studies, or small, single-institution, observational studies Results from a single institution often fail to identify a true relationship between outcome and risk factors because of the small sample size and short follow-up period Our analyses were based on the SEER database, which is considered to be the largest cancer registry Reports from a population-based cohort have the advantage of including many more patients, thus increasing the power to estimate the true effects of risk factors on survival Moreover, unlike most results from clinical studies, analysis of a population-based database includes not only those treated using formal protocols, but also those excluded from protocols because of comorbidity, tumor stage, or Yang et al BMC Cancer 2014, 14:654 http://www.biomedcentral.com/1471-2407/14/654 Page of Table Patient demographics and overall survival All patients Years OS (%) 10 Years OS (%) Characteristics No Events Rate 95%CI Rate 95%CI Entire cohort 1679 543 64.5 62.1-67.1 61.8 59.2-64.5 0-4 639 173 71.3 67.5-75.2 69.1 65.2-73.2 5-9 390 97 73.2 68.5-78.2 68.8 63.6-74.5 10-14 346 134 56.4 50.9-62.6 52.4 46.6-59.0 15-19 304 139 47.9 42.1-54.6 47.3 41.5-54.0 Age (years) p